Summary

• Fixes a regression introduced in commit 072165b ("Clean up OperationValidator: remove dead code and fix warnings") where the null-handling logic in requireSameOutputTypeShape() was incorrectly simplified
• Adds a test that reproduces the issue: fragments on unresolvable types producing spurious FieldsConflict errors like 'field/field/field' : returns different types 'Abstract' and 'null'
• The benchmarkDeepAbstractConcrete JMH benchmark in OverlappingFieldValidationPerformance was crashing due to this bug

Root cause

In OperationValidator.requireSameOutputTypeShape(), the null check was incorrectly simplified from:

if (typeA == null || typeB == null) {
    if (typeA != typeB) {
        return mkNotSameTypeError(path, fields, typeA, typeB);
    }
    continue;
}

to:

if (typeB == null) {
    return mkNotSameTypeError(path, fields, typeA, typeB);
}

The old code treated two null types as compatible (skip comparison). The new code unconditionally reported a conflict when typeB is null, even when typeA is also null or when null simply means the parent type was unresolvable.

Fix

Restored the original null-handling logic: when both types are null, they are treated as compatible and we continue to the next field. Only when one is null and the other isn't do we report a conflict.

Test plan

• New test "fields with null types from unresolvable parent should not produce spurious conflicts" passes
• All 33 tests in OverlappingFieldsCanBeMergedTest pass (0 failures)

https://claude.ai/code/session_01XYg9Dqneb941aStgKaZ1Em